This repository contains the code and data for the following paper:
Whose Hands Are These? Hand Detection and Hand-Body Association in the Wild (CVPR 2022)
We develop a method to detect hands and their corresponding body locations. Hands and bodies belonging to the same person have bounding boxes in the same color and identification numbers
Follow the installation instructions in INSTALL.md.
Please see the project page for the required dataset and annotation format.
Download models and place them in ./models/
.
Use the following command for training:
python train_net.py --config-file configs/BodyHands.yaml
Use the following command for evaluation:
python train_net.py --config-file configs/BodyHands.yaml --eval-only MODEL.WEIGHTS <path to model weights>
Use the following command to detect hands and their corresponding person locations:
python demo.py --input <path to a directory containing images> --thresh 0.7
Use the parameter thresh
to vary the precision and recall of detections.
The output images with hand detections and corresponding body locations' visualizations will be stored in ./demoOutput/
.
If you find our code or dataset useful, please cite our work using the following:
@inproceedings{bodyhands_2022,
title={Whose Hands Are These? Hand Detection and Hand-Body Association in the Wild},
author={Supreeth Narasimhaswamy and Thanh Nguyen and Mingzhen Huang and Minh Hoai},
booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
year={2022},
}
@inproceedings{handler_2022,
title={Forward Propagation, Backward Regression and Pose Association for Hand Tracking in the Wild},
author={Mingzhen Huang and Supreeth Narasimhaswamy and Saif Vazir and Haibin Ling and Minh Hoai},
booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
year={2022},
}
@inproceedings{contacthands_2020,
title={Detecting Hands and Recognizing Physical Contact in the Wild},
author={Supreeth Narasimhaswamy and Trung Nguyen and Minh Hoai},
booktitle={Advances in Neural Information Processing Systems (NeurIPS)},
year={2020},
}